Does Public Expenditure on Education Improve Well-Being? International Evidence

Kyklos, 2025; 0:1–17 https://doi.org/10.1111/kykl.12466 1 of 17 Kyklos ORIGINAL ARTICLE OPEN ACCESS Does Public Expenditure on Education Improve WellBeing? International Evidence RubénRubio-Ortiz | DavidPatiño | FranciscoGómez-García Departamento de Economía e Historia Económica, Facultad de Ciencias Económicas y Empresariales, Universidad of Sevilla, Sevilla, Andalusia,Spain Correspondence: David Patiño ([email protected]) Received: 5 September 2024 | Revised: 17 February 2025 | Accepted: 4 May 2025 Keywords: economics of education| public expenditure| subjective wellbeing ABSTRACT The aim of this article is to quantify the relationship between public expenditure on education and individual subjective wellbeing, providing empirical evidence of the social return on this investment. We use microdata from the European Social Survey (ESS) merged with macroeconomic variables from official sources. Econometric estimations are carried out using multilevel models. Our results show a positive association between public expenditure on education and individual wellbeing. However, this effect is not homogeneous across educational levels, as robust evidence of a positive contribution is only found for tertiary education. Furthermore, we explore whether this relationship is contingent on individuals' ideological preferences. Our findings indicate that individuals who espouse a conservative ideology exhibit a weaker effect compared to those with a progressive mindset. Nevertheless, the positive correlation between education expenditure and wellbeing persists for the latter group. To assess the robustness of our results, we have replicated the calculations using a different survey, specifically the Eurobarometer, and conducted estimations with alternative methodologies, which confirm their consistency. JEL Classification: H52, I26, I38, I31, I22 1 | Introduction Viewing education as an essential good for people's development and wellbeing (Sen1999) as well as for economic growth (Becker 1964; Schultz 1961; Solow 1957) makes investment in such a good of national interest and one of the central aspects of the policy of any country, as shown by its inclusion in the Sustainable Development Goals of the UN Agenda 2030. Receiving an adequate stock of education is a fundamental aspect of any person's life and its provision is a central element in any country's policy. Consequently, allocation of government expenditure towards education is significant, accounting for approximately 10.1% of total public expenditure in European Union countries for 2019 (Eurostat2023). Given the significance of this expenditure, investigating the returns on this investment is worthwhile. Return on this investment, mainly from a financial perspective, has been extensively studied from the classical approach of human capital theories—particularly through the educational return rate (Angrist and Krueger1999; Card1999; Mincer1974; Psacharopoulos and Patrinos2018). However, calculating educational return rates proves complex and has by no means been exempt from methodological concerns. This complexity arises, among other reasons, from the difficulties involved in quantifying certain nonmonetary benefits of education in monetary terms. As a result, these benefits are omitted due to the inherent difficulty of measuring them in monetary terms, among other limitations associated with traditional monetary measures (Stiglitz etal.2009). From another perspective, the presence of positive externalities at all levels also makes it difficult to quantify the full benefits of education. Thus, the relationship between education and subjective wellbeing can be analyzed This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2025 The Author(s). Kyklos published by John Wiley & Sons Ltd. 2 of 17 Kyklos, 2025 under two approaches (Leite etal.2024). At the microeconomic level, this paper finds a positive but weak relationship. However, at the macroeconomic level, the relationship is also positive, but very strong. In this context, our work, in the exercise of evaluating the effect of public educational spending on subjective wellbeing, allows us to capture the positive externality that it generates and that can explain this relationship, which is not possible to capture in studies that are limited to the individual level. The subjective wellbeing approach offers a good alternative to address these and others limitations. This approach provides a framework to estimate individual's overall utility across various dimensions from a holistic perspective, taking into account factors such as income and wealth, inequality in its distribution, environmental concerns, leisure, and the sociocultural context, among others. This allows us to identify actions, contexts, and policies that help to enhance people's wellbeing and makes it feasible to estimate a return rate of education that includes all its effects. This is achieved by monitoring individuals' selfreported life satisfaction or utility. By collecting such reports from a diverse range of individuals—at different points in time and across various administrative units—it is possible to infer relationships between contextual circumstances and the varying degrees of utility experienced by individuals (Frey and Stutzer2002; Odermatt and Stutzer2017). In order to clarify the conceptual framework underpinning this research, it is important to consider the distinction between the three dimensions of the key independent variable of this paper, subjective wellbeing: evaluative, hedonic, and eudaimonic (Graham and Nikolova2015). First, evaluative wellbeing (life satisfaction) refers to a respondent's assessment of the degree to which they are satisfied with their life as a whole. Secondly, hedonic wellbeing refers to the emotional and affective component of subjective wellbeing. It reflects how people experience their lives as opposed to how they value it generally (Kahneman and Krueger2006). Finally, eudaimonic wellbeing reflects the Aristotelian notion of happiness as a life purpose and focuses on the flourishing and fulfillment of human potential (Organisation for Economic Cooperation and Development 2013). On the other hand, Becchetti etal.(2023) find a relationship between eudaimonic wellbeing, which they associate with the sense of life, and subjective survival probability, which they use as a proxy for selfassessed life expectancy. In this way, they link this subjective measure of wellbeing with survival probability, which becomes particularly relevant in individuals' decisions regarding aspects such as consumption, savings, and retirement. Given that full life satisfaction is one of the objectives to be achieved for individuals and, by extension, governments—given their commitment to the wellbeing of their citizens (Oishi and Diener2014; Pereira etal.2023; Schubert2012)—exploring in depth the policies and actions that help achieve this objective proves crucial. In addition, being public financial resources limited, it is important to allocate spending appropriately among different policies or spending programs and setting priorities in defining the policies to be implemented. This paper seeks to accomplish such a goal. We evaluate the returns of public education; however, in contrast to the previous literature, we assess it as through the subjective wellbeing approach. The results obtained offer interest in themselves as they allow for a different, and also enriching, way of assessing public policies, and the findings complement the traditional assessment derived from classical human capital theories. The main objective of this paper is to assess the effect of public spending on education. To do so, we propose to quantify it through the wellbeing generated by the expending. Using reported life satisfaction collected through surveys that we employ as a proxy for wellbeing. However, our a priori approach is that public spending on education improves people's lives and, therefore, has a positive return in terms of wellbeing. Thus, our paper aims to empirically test the positive effect of public education on individuals' selfreported wellbeing and, therefore, generates improvements in the wellbeing of societies. However, we also consider a priori that, although positive, the return of public spending on education on subjective wellbeing is not homogeneous across educational levels (primary, secondary, and tertiary), in line with its contribution to financial performance. Our empirical work also aims to quantify this effect for the different educational levels. Finally, we also consider the extent to which people's political preferences modify the effect of public spending on the extent to which public education produces the effect we hypothesize on subjective wellbeing, as it is possible that this effect is mediated by this aspect. To the best of our knowledge, our study contributes to the literature by testing in a novel way whether the effects of education spending are similar across educational levels and whether political preferences modify the effects on individual welfare. Our study also raises concerns about the robustness of our results. We validate our findings using two different surveys that use different response scales for the endogenous variable. In addition to the main objective, the paper also has secondary objectives that represent outstanding contributions to the previous literature. In particular, the paper conducts a literature review on the relationship between subjective wellbeing and education, distinguishing between the individual and social approaches. Our paper focuses primarily on the social approach. For this approach, there is a small literature review. While the majority of prior studies have predominantly adopted an individual approach to examining the relationship between education and subjective wellbeing (among others, Araki2022; Clark and Oswald1996; Frey and Stutzer2000; Kristoffersen2018), our paper adopts a social approach by combining individual variables from survey microdata with aggregate variables. We focus on public expenditure in education, as an investment, and use it as the main study variable to test its relationship with wellbeing. In this way, we emphasize the macro aspect of the policy but evaluate it in terms of the effects it has on individuals. The results confirm the initial hypotheses. Public expenditure on education has a positive rate of return in terms of subjective wellbeing, and this return is not uniform across levels of education. Specifically, there is strong evidence of a positive return only for public expenditure on tertiary education. Furthermore, we also find that some individual characteristics play a modifying role. It underlines the fact that individuals who display conservative political preferences exhibit a smaller effect of public investment in education compared to other subsets of the population with more progressive mindset. 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 3 of 17 The paper is structured as follows. In the next chapter, we provide a brief literature review, followed by the presentation of the theoretical framework and the data used in the empirical study. The results are then presented and discussed. Finally, we draw conclusions and suggest promising avenues for future research. In addition, we include the bibliography used and appendices for supplementary information. 2 | Literature Review The link between subjective wellbeing and education has mainly been studied from an individual approach. Numerous articles have explored what effect individuals' level of education—sometimes measured as years of schooling and on other occasions in terms of the maximum level achieved—has on subjective wellbeing (see Bücker etal.2018; Tan etal.2020; Witter etal.1984). Many of these studies confirm the positive relationship that the known education benefits anticipate (see, among others, Blanchflower and Oswald2004; Easterlin2001; FerreriCarbonell2005; Frey and Stutzer2000). However, as yet no consensus has been reached, because several articles report a negative or nonsignificant relationship. One common explanation for this negative relationship involves the unfulfilled expectations of individuals with higher education (Clark and Oswald 1996; Kristoffersen 2018). Other possible causes have also been identified, such as stress and time constraints resulting from highly skilled jobs (Nikolaev2018), the need to spend everincreasing amounts of time and money on formal education in a more competitive society (Veenhoven and Berg2013), the opportunity cost of education and its perceptibility (Ferrante2009), or a tendency among unhappy individuals to pursue higher levels of education (Veenhoven2010). However, other studies seek to explain these results without calling into question the positive contribution made by education. Rather than only estimating the direct effect—which they assume to be null— they attempt to estimate the indirect effect, which becomes blurred among other variables. Other studies involve the use of estimation techniques that allow a nonlinear relationship to be studied (Cuñado and Pérez de Gracia2012; Hartog and Oosterbeek1998; Helliwell2003; Powdthavee etal.2013). There is abundant amount of literature addressing the individuallevel approach, especially due to the inclusion of education level as a control variable in most subjective wellbeing models. In contrast, very few studies have explored the relationship between aggregate education variables and individual subjective wellbeing using a monographic approach. Despite the usefulness of such an approach for policy design, few studies have linked people's wellbeing with the variables that define education policies and mainly with spending on education. Overall, the results found in the literature allow us to build our starting hypothesis that public education can be considered a powerful instrument to raise the welfare levels of individuals. Among the few existing studies, Bukenya etal.(2003) include public expenditure on education in 1988 for certain regions of West Virginia (United States) as a variable for explaining life satisfaction. The authors find its impact to be positive and significant, yet also find local public expenditure on education to be nonsignificant. Similarly, exploring a broader dimension beyond public spending on education, AcuñaDuarte etal.(2024) find that preferences for reforming the Chilean Constitution can be driven by greater discontent with the provision of public goods. Hessami (2010) and Ho and Ng (2016) offer more closely related references because they link subjective wellbeing with size of government. The former study employs certain econometric models for 12 European Union countries, using public expenditure on education as an exogenous variable in many of them. The author's findings suggest an inverse Ushaped relationship, indicating an optimal level of public expenditure that maximizes subjective wellbeing. Building upon this notion, Ho and Ng(2016) explore the optimal level of public expenditure in terms of maximizing subjective wellbeing using panel data from 78 countries. The authors identify optimal government size as ranging between 35.3% and 37.8% of gross domestic product (GDP), with the optimal level of public expenditure on education being between 3.96% and 4.25% of GDP. Cheung and Chan(2011) conduct an econometric analysis using aggregated and averaged data at country level to estimate what impact public expenditure on education—measured as a percentage of GDP—has on happiness. Using twotime references of public expenditure on education to explain average happiness between 2000 and 2008, they found that public expenditure on education in 2004 had a positive effect on average life satisfaction. However, they found no statistically significant relationship for 2001. Also using average countrylevel data, Kim and Kim(2012) find a positive effect of spending in education on subjective wellbeing. Using World Values Survey (WVS) microdata for a panel of 59 countries between 1981 and 2013, Jun(2015) applies multilevel regressions with robust standard errors with random intercept and coefficients that include individuallevel and aggregated variables. The exogenous variables include a composite variable—generated by principal component analysis—which includes public expenditure on education, public social expenditure, the Gini index, and total secondary education enrollment. The author finds an ambiguous relationship between this composite variable and subjective wellbeing and concludes that no clear effect could be determined. Using individuallevel data for 109 countries from 1998 to 2008, Ortega Gil(2021) applies fixedeffects models to estimate a positive relationship between public expenditure on education and life satisfaction. Qasim and Grimes(2022) examine changes in life satisfaction between two rounds. They use a dynamic OLS model, where they introduce macroeconomic variables referenced to the first round to explain changes in life satisfaction in the period between the two rounds. Using public expenditure on education as one of the exogenous variables, they find no significant effect on life satisfaction. Finally, Ortuzar etal.(2021) employ singlecountry data with a regional approach to quantify the effect of public expenditure in health on subjective wellbeing. Their model includes public expenditure on education as a control variable and evidences a 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 4 of 17 Kyklos, 2025 positive and significant coefficient. The study is notable for its inclusion of individuals' political preferences and because it examines their interaction with the health expenditure variable. The findings highlight that the impact of this variable on life satisfaction is contingent upon political ideology. 3 | Theoretical Framework and Econometric Strategy This paper is grounded on the premise that life satisfaction revealed by individuals through surveys serves as an accurate proxy variable for individual utility levels. The justification for employing this method rests on the assumption that individuals possess the most accurate knowledge of their own utility, thereby making direct inquiry the optimal strategy (Frey and Stutzer2002). This approach signals a return to classical utilitarian ideas of utility measurability put forward by authors such as Bentham or Edgeworth (AnsaEceiza and GómezGarcía 2019). This theoretical corpus was relegated by the revealed preferences approach, which has the advantage of only requiring preferences to be ordered rather than requiring a measure of to what extent one choice is preferred over another. However, the revealed preferences approach makes unrealistic assumptions such as the extreme rationality of individuals, which has been subject to considerable debate (Kahneman etal.1997; Kahneman and Sugden2005). Given these limitations, behavioral economists have developed a robust theoretical framework that provides a solid foundation for empirical studies on subjective wellbeing. The model employed in this study is based on the life satisfaction approach (Frey etal.2010), which enables the net effect of different government policies on individual wellbeing to be estimated. This approach has the advantage of avoiding some traditional limitations such as strategic response behavior because—by asking about life satisfaction in a wide sense—it prevents responses directly linked to policy objectives (Kahneman and Sugden2005; Odermatt and Stutzer2017). Moreover, the comprehensive and integrated perspective of the approach allows policy consequences that are often overlooked by other methodologies to be explored. Specifically, in addition to the direct effects, the estimated effects encompass externalities, unobserved effects, or indirect channels of transmission. This comprehensive assessment is possible because individuals value their lives as a whole, considering future projections, past experiences, and relative circumstances in comparison to others (Frey and Stutzer2002; Andriani and Ashyrov 2022). From a broad perspective, individuals' level of wellbeing or real utility ( Uijt ) depends on certain individual characteristics ( Xijt ) as well as aggregate, social, or contextual variables ( Yjt ), one of which is our main exogenous variable—public expenditure on education ( EDUjt ): where the subscripts represent the individual (i), the country where they live (j), and the time (t). This utility function can be approximated with the life satisfaction variable, as a proxy, including an error term which—in addition to random perturbation—reflects individuals' inherent inability to accurately report their true life satisfaction level, among other circumstances (Blanchflower and Oswald 2004; Ortuzar etal.2021): Similar to numerous prior studies (such as Di Tella etal.2003; Patiño etal.2022), we employ individuallevel data from survey microdata, together with aggregated countrylevel data. However, unlike previous studies that group individual data using averages, which allows them to make estimations using simple econometric techniques, we exploit the richness of working with individual microdata to operate with two levels. As pointed out by Hox etal.(2010), relying solely on individual data averages may lead to a loss of accuracy and yield different results compared to estimates based on microdata. By using a nested data structure, we gain flexibility, better control over heterogeneity through individual variables, and the ability to employ more techniques such as variable interactions. Furthermore, this approach facilitates a better understanding of the relationships between variables at different levels. Nonetheless, the nested data structure does pose one challenge because it violates the assumption of independence, thereby possibly resulting in biased estimators and excessive Type I error, that is, it has a high tendency to estimate false positives. To address this issue, we employ multilevel models (Gelman etal.2012). We estimate multilevel models based on this theoretical framework. This type of model, which is also known as hierarchical, mixed, or nested, is a combination of fixedeffects and randomeffects models and is very useful in contexts where observations are nested in clusters. This method allows interactions across levels to be explored, thereby enabling analysis of how macroeconomic policies and contexts influence individual attitudes, thoughts, and behaviors (Fairbrother2014). Our choice of this strategy diverges from common alternatives in the literature that have previously explored this research question. One frequently employed technique is OLS regression with fixed effects, where dummy variables are employed to control for spatial or temporal heterogeneity. However, this approach runs the risk of dummies absorbing a substantial amount of level variability, possibly leading to multicollinearity issues with exogenous variables at the country–year level (Qasim and Grimes2022). Multilevel estimation avoids this situation. Choice of group is crucial in multilevel models. In our case, data are clearly grouped by countries. Nevertheless, because we draw on different survey rounds, these rounds may possess timerelated aspects that are shared across yearly observations. This grouping highlights the fact that from institutionalstructural aspects, determined by the country in which the individual respondent resides, to economic aspects, established by the year to which the response corresponds, all observations that correspond to the same country–year cluster have common relationships that correlate them. Some studies have emphasized that the minimum number of clusters to estimate errors (1) Uijt( X ijt ,Y jt ,EDU jt), (2) Lifesatijt =U ijt( X ijt ,Y jt ,EDU jt) +𝜀 i. 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 5 of 17 accurately must be at least between 30 and 50 and that the number of clusters is more important than the number of observations in each cluster (Maas and Hox2005; Paccagnella2011). In view of this, we consider observations clustered by country–year groups. This cluster structure increases the number of higher level units while allowing us to capture unobserved heterogeneity related to spatial and temporal dimensions without the difficulties caused by a less precise method such as simply including fixed effects. Thus, our multilevel model is structured in two. Our second level (or higher level) comprises variables at the country–year group level, while the first level (or lower level) consists of individual observations, which, statistically freed from the structural and conjuncture effect, can adequately determine the effect of public education, the variable we intend to measure. As regards the mixed models employed, we treat the intercept as random while considering the coefficients as fixed. The group intercept serves as the starting point for individual regressions within each country–year group. This modeling approach enhances interpretability, as the coefficients obtained indicate the effect of exogenous variables on the formation of the group intercept. Specifically, a positive coefficient signifies that individuals belonging to the same country–year group have a greater likelihood of attaining higher levels of life satisfaction because of variations in the level of the variable studied. The general equation of the twolevel mixed model estimated is as follows: Level 1: individuals in the survey Level 2: country–year groups Combined: where ij represents an individual i within country–year group j, LSij denotes their life satisfaction level, Xij represents the vector of individual characteristics that vary within the country–year group, and Mj is a vector of macroeconomic variables that remain constant within each group but vary across groups. The group random error term ( uj ) and individual error term ( 𝜀ij ) complete the general equation. Given the ordinal nature of the endogenous variable—life satisfaction—it is common practice to estimate discrete choice models such as the ordinal logit model. However, previous literature suggests that linear models may yield similar results to discrete choice models when the discrete life satisfaction variable employs a highresponse scale (FerreriCarbonell and Frijters2004; Kristoffersen2018). The cardinality assumption is generally accepted in such cases (Kristoffersen2017). In our main study, given that it provides a high scale of responses to the life satisfaction variable, we employ linear models following this logic, as they facilitate the interpretation of coefficients, which is already complex in the case of multilevel models. Additionally, this approach helps reduce the computational complexity required by the algorithm, which is particularly demanding for ordinal models. We estimate ordinal models as a robustness check. Drawing on the European Social Survey (ESS) dataset—which provides a highresponse scale for the life satisfaction variable— we employ both linear mixed models and ordinal logit mixed models as a robustness check. For the Eurobarometer dataset— which contains a life satisfaction variable with only four categorical levels—we only employ the ordinal mixed logit model. We estimate models using R, Version 4.2.2. Specifically, we use the lme4 package for linear mixed regressions (Bates etal.2015) and the ordinal package for ordinal logit mixed regression (Christensen2022). 4 | Data We use the microdata from the ESS (ESS ERIC2020), which contains all the individual data we require. Descriptive statistics for individual control variables are contained in Table 1. Similarly, the descriptive statistics for macroeconomic variables, (3) LSij = 𝛾i + 𝛽ij ∗ Xij + 𝜀ij. (4) 𝛾i = 𝛾0 + 𝛽j ∗ Mj + uj. (5) LSij = 𝛾 0+ 𝛽ij ∗ Xij + 𝛽j ∗ Mj + uj + 𝜀ij, TABLE 1 | Descriptive statistics of the European Social Survey variables. Variable N. Obs. Mean Std. Dev. Min. Max. Life satisfaction 336,521 7.02 2.2 010 Age 336,521 48.43 18.26 14 114 Political preferences 336,521 1.97 0.75 1 3 Marital status 336,521 2.56 1.78 1 5 Unemployed 336,521 0.05 0.22 0 1 Education level 336,521 1.96 0.65 1 3 Individual health (auto evaluation) 336,521 2.19 0.91 1 5 Sex 336,521 1.52 0.5 1 2 Immigrant 336,521 1.09 0.29 1 2 Source: Own elaboration based on data used (ESS ERIC2020). 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 6 of 17 Kyklos, 2025 indicating variability, are presented in Table 2. The specific rounds, years, and countries employed in the survey are detailed in TableA1. Macroeconomic variables are merged with individual observations using country and year as reference, which sometimes, for a number of reasons detailed below, differs from survey round edition year. We utilize data from the nine rounds that have been conducted, spanning the period 2002–2020. The period covered is sufficiently long to capture variations in public education, which constitutes a structural policy in European countries and, therefore, exhibits very limited variability. Additionally, this period is of particular interest as it includes numerous modifications and significant historical changes that make it a valuable case study. Specifically, it encompasses important structural transformations, particularly in some countries in Eastern Europe, which have undergone profound changes in their situation. It also includes various economic conjunctures that have significantly affected the provision of public services, and education in particular, providing a significant degree of variability, in addition to that generated by the different national models employed. In particular, it covers the strong economic growth of the early 21st century, the Great Recession of 2008–2012, and the subsequent recovery. The recession represented an interesting asymmetric shock that particularly affected Southern European countries. We have focused our work on the ESS, discarding other sources such as the Eurobarometer because it is important to note that the measurement of life satisfaction—which constitutes the endogenous variable—differs between the two surveys, making it difficult to compare the results directly. Specifically, the ESS measures life satisfaction through a categorical variable ranging from 0 (extremely unsatisfied) to 10 (extremely satisfied), while the Eurobarometer limits the possible values with four levels (very unsatisfied, unsatisfied, satisfied, and very satisfied). In this regard, although the latter dataset also contained information on the variables we required, we decided to exclude it due to the challenges in comparing both results. We limited the use of this source to conducting a robustness check on the qualitative results. The estimated models incorporate individual control variables commonly used in previous studies, such as age, educational level, gender, household size, occupation, and political preferences (Radcliff2013). Continuous variables have been normalized and centered using the mean—following the approach of Bell etal.(2018). This normalization facilitates the calculation in the likelihood function optimization process. Similarly, the macroeconomic variables employed align with those used in previous studies within this field (see, among others, Hessami2010; Ortega Gil2021; Ortuzar etal.2021). From it, we intend to control for macroeconomic aspects and institutional characteristics, which have to be captured in a specific way. The macroeconomic variables included in the models do not have a high correlation between them, and there is no individual correlation greater than 0.7. In addition, the VIF test is well below 5, such that multicollinearity is unlikely. Finally, to prove that the mixed model is a suitable technique, we add a base linear mixed model to obtain the ICC metric. The intraclass correlation coefficient (ICC) test captures the percentage of variance that can be explained by differences between units at higher level(s), and it is ranged between 0 and 1. If the ICC test is close to 0, a multilevel model is not necessary. As the value increases, there is a greater violation of the assumption of independence, serving as an indicator of the need to use a multilevel model. In social research, it is common for the ICC to range between 0.05 and 0.20 (Peugh2010). Public expenditure on education variables is taken from the World Bank, with the original source being UNESCO. Additionally, TABLE 2 | Descriptive statistics of the macroeconomic variables employed in the analysis. European Social Survey N. Obs. Mean Std. Dev. Min. Max Total public expenditure on education, % GDP 364 5.37 1.15 3.3 8.58 Public expenditure on primary education per student, % GDP per capita 217 20.97 4.01 10.79 32.35 Public expenditure on secondary education per student, % GDP per capita 217 24.89 4.83 14.33 38.33 Public expenditure on tertiary education per student, % GDP per capita 217 31.39 10.66 13.29 71.97 Unemployment rate 364 7.60 3.59 2.02 26.09 Inflation rate 364 2.22 2.27 −4.45 15.88 Gross national income per capita 364 36,117.83 21,149.14 1540.0 105,070.0 Public expenditure on general goods and services, % GDP 364 6.17 1.92 2.67 13.04 Public expenditure on health, % GDP 364 6.17 1.65 1.63 8.87 Public expenditure on culture, % GDP 364 1.28 0.46 0.3 3.46 Source: International Monetary Fund(2024) and World Bank(2024). 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 7 of 17 macroeconomic variables such as inflation rate, unemployment rate, and income per capita are obtained from the World Bank. Data related to other public expenditures are sourced from the COFOG database of the International Monetary Fund, specifically, public expenditure on general goods and services (GF01), health (GF09), and culture (GF08), all expressed as a percentage of GDP. Total public expenditure on education is relativized as a percentage of GDP. Conversely, public expenditure on education disaggregated by education levels (primary, secondary, and tertiary) is divided by the number of students enrolled in each level and subsequently expressed as a percentage of GDP per capita. The reason for not dividing total public expenditure on education by the number of students enrolled is to prevent loss of observations due to the unavailability of enrollment data for certain levels, which would hinder obtaining the total number of students enrolled. This transformation method allows for a more accurate measurement of the country's investment effort by taking into account its wealth. 5 | Results and Discussion This section presents the results that have been obtained from the different regressions performed. The main results are presented in the subsequent subsections based on the research question tested and clearly show the positive impact of public spending on education on life satisfaction. In general, the different regressions have estimated coefficients showing the expected signs in line with previous studies (Dolan etal. 2008; Jun 2015; Venetoklis 2019), while they are also the subject of commentary in the text. The coefficients for the individual and macroeconomic control variables used are detailed in TableA2. 5.1 | The Impact of Total Public Expenditure in Education on WellBeing The results regarding the impact of public expenditure in education—expressed as a percentage of GDP—on life satisfaction are presented in Table3. The information is organized in columns corresponding to the type of model used. Considering the qualitative nature of the endogenous variables, we employed mixed ordinal logit regressions. However, because the variable of life satisfaction consists of 11 response levels, we also use a linear mixed model. This provides an alternative interpretation based on cardinality and serves as a robustness check for the results obtained. As shown in the table, the results demonstrate a clear positive relationship between public expenditure on education and life satisfaction, showing that public spending on education appears to be a useful tool for raising individuals' life satisfaction. As shown, the relationship is statistically significant in all estimated models. The robustness of the findings is remarkably high, indicating a consistent relationship across different perspectives of the endogenous variable. In particular, the statistically significant relationship is found in the regressions performed holds for both estimation methods, cardinal and ordinal. All estimated models find the estimator to be statistically significant with a positive sign, providing compelling evidence that investing in education is an effective means of enhancing people's wellbeing. This confirms our main research question that public spending on education has a positive impact on life satisfaction and is a useful instrument to increase it. These results are in consistent with previous studies (Hessami2010; Kim and Kim2012; Ortega Gil2021; Ortuzar etal.2021) that had indirectly found this relationship. We add the base linear mixed model, as Model 1, to prove that the statistician technique is suitable for this data. TABLE 3 | Effect of public expenditure on education on subjective wellbeing in European countries (2002–2020). Endogenous variable: Life satisfaction (0–10) Linear mixed model Ordinal logit mixed model (1) (2) (5) Individual control variables No Yes Yes Macroeconomic control variables No Yes Yes Public expenditure on education (% GDP) 0.082** 0.108*** (0.034) (0.032) Std. Dev. (group) 0.906 0.464 0.430 Std. Dev. (observations) 2.113 1.866 Num. Obs. 428,508 336,521 336,521 Num. Groups 392 364 364 AIC 1,859,195.6 1,376,253.5 1,284,524.3 BIC 1,859,228.5 1,376,543.1 1,284,899.8 ICC 0.16 Note: Standard errors in parentheses, observations nested at country–year level. Source: Own elaboration based on data used (ESS ERIC2020; International Monetary Fund2024; World Bank2024). **p < 0.05. ***p < 0.01. 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 8 of 17 Kyklos, 2025 In addition, we can estimate the quantitative effect using the coefficient values. Quantitative interpretation should be taken with caution, given the nature of the life satisfaction variable which, even in the survey that allows for a larger number of responses, is always mainly ordinal in nature. Specifically, the coefficient value indicates the magnitude of the change in welfare that occurs when public expenditure on education varies by one standard unit of GDP, taking into account the previous standardization. The linear model reveals that a onestandardunit increase in public expenditure on education corresponds to a variation of 0.082 points in life satisfaction among respondents. Interpreted in terms of variation raw GDP points (not standard units), this result translates into an increase in life satisfaction of between 0.015 and 0.149 for each percentage point increase in education spending, with a 95% confidence interval. Ordinal logit mixed model coefficients represent the change in the logodds of being in a higher category of life satisfaction for a one standard deviation increase in the public expenditure on education variable, holding all other variables constant. When exponentiated, this corresponds to an odds ratio of 1.1135, indicating that a one standard deviation increase in public education expenditure is associated with an 11.35% increase in the cumulative odds of reporting higher life satisfaction. In terms of marginal effects, a one standard unit increase in educational expenditure consistently decrease the probability of reporting low and medium satisfaction levels (0–7), while increasing the probability of reporting high satisfaction levels (8–10). Specifically, when evaluated at the mean, this increase reduces the probability of reporting medium satisfaction (Level 5) by 0.81 percentage points, while increasing the probability of reporting high satisfaction levels by 0.99 and 1.02 percentage points for Levels 8 and 9, respectively, with a 0.68 percentage point increase for the maximum level. 5.2 | Effects of Public Expenditure on Education on WellBeing by Level of Education The next question we analyze concerns the quantification of the impact of public expenditure on education, distinguishing among the levels of education at which the money is allocated (primary, secondary, and tertiary education). Thus, we quantify the extent to which investments in public education are effective in generating welfare. In order to disaggregate public education expenditure by level, and given the absence of published data, public expenditure on education has been adjusted relative to the number of students enrolled in each educational level and expressed as a percentage of GDP per capita. Results from the estimations are presented in Table4. As in the previous subsection, we offer separate information based on the different estimation techniques employed. The econometric models used in this analysis are similar to those previously described, and we report only the coefficients of key variables. The details of the other results, the coefficients of the control variables, are provided in TableA2. These findings show that while public expenditure on education has a positive effect on subjective wellbeing, this effect is not generated across educational levels. In particular, the only level that consistently produces significant and positive results is tertiary education. We can therefore conclude that investment in higher education is the only type of investment that has a positive effect on wellbeing. In contrast, the other two levels of education considered either have no effect or even reduce it. Given the lack of previous studies addressing this specific question that could serve as a basis for comparison, it would be interesting to compare these findings with estimates of societal and individual return rates. In general, estimated return rates tend to be more uniform and even decrease with higher levels of education (Psacharopoulos and Patrinos2018). Conversely, the situation is different when assessing the impact on wellbeing, as only higher levels of education exhibit a clear influence. In the case of public expenditure on primary and secondary education, the coefficients are not significant in any model, showing that this level of education has no effect on wellbeing. In the light of the results obtained, it can be seen that the positive effect of public education on welfare is limited to higher education, which seems to be the real instrument that produces this result. On the other hand, the other levels of education either have no effect at all. Obviously, such a counterintuitive result is both surprising and complicated to interpret. TABLE 4 | Public expenditure on education differencing by education levels effect on subjective wellbeing. Endogenous variable: Life satisfaction (0–10) Linear mixed model Ordinal logit mixed model (3) (6) Individual control variables Yes Yes Macroeconomic control variables Yes Yes Primary, public expenditure on education by student (% GDP per capita) 0.018 0.042 (0.046) (0.042) Secondary, public expenditure on education by student (% GDP per capita) −0.060 −0.072 (0.049) (0.044) Tertiary, public expenditure on education by student (% GDP per capita) 0.159*** 0.180*** (0.053) (0.048) Std. Dev. (group) 0.516 0.467 Std. Dev. (observations) 1.871 Num. Obs. 211,320 211,320 Num. Groups 217 217 AIC 865,402.3 807,681.8 BIC 865,699.8 808,061.4 Note: Standard errors in parentheses, observations nested at country–year level. Source: Own elaboration based on data used (ESS ERIC2020; International Monetary Fund2024; World Bank2024). ***p < 0.01. 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 9 of 17 On the one hand, a possible explanation for this result could be the sample of countries to which we restricted the work. In particular, given that European countries are the region in which the highest shares of public expenditure on education have been advanced, the impact of public expenditure on education at primary and secondary levels may indicate a saturation of public expenditure. With primary and secondary education having been universalized with relatively high standards of quality in European countries, one additional unit of expenditure may not have a significant effect and can even become negative if sections of the population prefer other forms of education. Against this background, the data show that investment in higher education is quite productive in generating welfare, possibly because there are still widespread shortages and deficiencies and only part of the population can afford this type of educational service. In addition, higher education can generate important social returns in terms of externalities, through the labor market or through R&D&I, knowledge transfer, and spillover effects produced by universities. The result may therefore reflect this situation. This implies that in societies where most people have primary or secondary education, differentiation is mainly through higher education. It is reasonable to assume that enhancing the quality or increasing the quantity of higher education could have a greater impact on individual wellbeing, particularly when the supply is remains relatively low compared to total demand. In such cases, public investment may be perceived as highly costeffective in terms of welfare. On the other hand, the limited contribution of public expenditure on basic and secondary education to subjective wellbeing could be interpreted as individuals not fully accounting for the significant intertemporal effects of public education spending. In reality, increases in public funding for education require considerable time to materialize into tangible educational benefits that individuals can perceive. In societies such as those in Europe, where public education at lower levels is fully established, this expenditure may be taken for granted, and any increases are likely aimed at enhancing service quality. However, such improvements may not lead to immediate wellbeing gains, as their benefits often become apparent only when the recipients reach adulthood and integrate into the labor market and society as a whole. Consequently, these expenditures may not generate shortterm wellbeing improvements and could even explain the negative sign if their funding has adversely affected other policies or certain groups. Both interpretations are not mutually exclusive and can be seen as complementary explanations of the coefficients we have estimated. 5.3 | How the Effect of Public Expenditure on Education Changes When Considering Political Preferences The third exercise that we carry out is to analyze whether individual political preferences shape what effect public expenditure in education has on subjective wellbeing. In this way, we explicitly take into account the possibility of significant heterogeneity between individuals when interpreting the effects of public education and the amounts spent on it. The concrete way in which we approach this analysis is by introducing a variable obtained by multiplying the main exogenous variable, public expenditure on education, by another variable defining the ideological preferences of individuals. The characterization of individuals according to their ideological preference variable is derived from the surveys themselves. Specifically, from a question in the survey, where individuals are asked to position themselves on a discrete scale of political ideology. In the ESS, it ranges from 0 to 10, with the lower end representing the political left and the higher end representing the political right. In order to ensure comparability between surveys and to facilitate interpretation and calculations, we have chosen to convert the initial responses into a threelevel variable: left, center, and right, which allow individuals to be characterized. Specifically, we adopt the following grouping: left if the individual positions themselves within the response range of (0–4), center (5–6), and right (7–10). Following the structure of the previous subsections, Table 5 presents the coefficients of key variables to address the research question. By introducing an interaction between the categorical variable of ideological preferences—with the center ideology as the base level—and public expenditure on education, we can calculate the effect of public education on welfare when judged by a person with different ideological preferences. In the light of the results obtained, we can conclude that although the effect of public education as an instrument for TABLE 5 | Effect of individual political preferences on satisfaction generated by public expenditure on education. Endogenous variable: Life satisfaction (0–10) Linear mixed model Ordinal logit mixed model (4) (7) Individual control variables Yes Yes Macroeconomic control variables Yes Yes Public expenditure on education (% GDP) 0.103*** 0.130*** (0.034) (0.032) Pub. Exp. Edu. * Left 0.006 0.009 (0.008) (0.007) Pub. Exp. Edu. * Right −0.081*** −0.067*** (0.008) (0.007) Std. Dev. (group) 0.465 0.435 Std. Dev. (observations) 1.865 Num. Obs. 336,521 336,521 Num. Groups 364 364 AIC 1,376,129.6 1,284,437.2 BIC 1,376,440.7 1,284,834.1 Note: Standard errors in parentheses, observations nested at country–year level. Source: Own elaboration based on data used (ESS ERIC2020; International Monetary Fund2024; World Bank2024). ***p < 0.01. 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 16 of 17 Kyklos, 2025 (1) (2) (3) (4) (5) (6) (7) Marital status—Divorced −0.563*** −0.579*** −0.564*** −0.517*** −0.523*** −0.517*** (0.012) (0.015) (0.012) (0.011) (0.014) (0.011) Marital status—Widowed −0.508*** −0.493*** −0.507*** −0.492*** −0.472*** −0.492*** (0.013) (0.017) (0.013) (0.013) (0.016) (0.013) Marital status—Never married −0.401*** −0.406*** −0.402*** −0.418*** −0.423*** −0.419*** (0.010) (0.012) (0.010) (0.009) (0.012) (0.009) Unemployed −0.989*** −0.974*** −0.989*** −0.856*** −0.832*** −0.856*** (0.015) (0.019) (0.015) (0.014) (0.018) (0.014) Education level—Secondary 0.136*** 0.137*** 0.139*** 0.099*** 0.106*** 0.100*** (0.014) (0.018) (0.014) (0.014) (0.017) (0.013) Education level—Tertiary 0.386*** 0.397*** 0.388*** 0.294*** 0.296*** 0.294*** (0.015) (0.020) (0.015) (0.015) (0.019) (0.015) Health—Good −0.507*** −0.511*** −0.508*** −0.591*** −0.591*** −0.591*** (0.009) (0.011) (0.009) (0.008) (0.010) (0.008) Health—Average −1.177*** −1.179*** −1.178*** −1.234*** −1.226*** −1.235*** (0.010) (0.013) (0.010) (0.010) (0.012) (0.010) Health—Bad −2.119*** −2.109*** −2.120*** −2.047*** −2.025*** −2.048*** (0.015) (0.019) (0.015) (0.015) (0.019) (0.015) Health—Very bad −3.084*** −3.103*** −3.084*** −2.858*** −2.866*** −2.859*** (0.030) (0.038) (0.030) (0.031) (0.039) (0.031) Sex—Woman 0.109*** 0.100*** 0.108*** 0.124*** 0.118*** 0.123*** (0.007) (0.008) (0.007) (0.006) (0.008) (0.006) Immigrant −0.168*** −0.191*** −0.169*** −0.169*** −0.191*** −0.170*** (0.012) (0.015) (0.012) (0.011) (0.014) (0.011) Political preferences—Left −0.184*** −0.188*** −0.187*** −0.173*** −0.179*** −0.175*** (0.008) (0.010) (0.008) (0.007) (0.009) (0.007) Political preferences—Right 0.309*** 0.316*** 0.318*** 0.321*** 0.325*** 0.328*** (0.008) (0.010) (0.008) (0.008) (0.010) (0.008) Note: Standard errors in parentheses, observations nested at country–year level. Source: Own elaboration based on data used (ESS ERIC2020; International Monetary Fund2024; World Bank2024). *p < 0.1. **p < 0.05. ***p < 0.01. TABLE A2 | (Continued) 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 17 of 17 TABLE A3 | Estimation results for the control variables from the Eurobarometer. (8) (9) (10) Threshold points: 1|2 −3.402*** −3.394*** −3.402*** (0.035) (0.046) (0.034) 2|3 −1.467*** −1.377*** −1.467*** (0.034) (0.045) (0.034) 3|4 1.725*** 1.828*** 1.725*** (0.034) (0.045) (0.034) Macroeconomic control variables Unemployment rate −0.206*** −0.064 −0.206*** (0.037) (0.047) (0.035) Inflation rate −0.098*** −0.145*** −0.098*** (0.033) (0.041) (0.033) GDP per capita 0.536*** 0.646*** 0.536*** (0.034) (0.045) (0.034) Public expenditure on general services (% GDP) −0.111*** −0.160*** −0.111*** (0.037) (0.049) (0.036) Public expenditure on health (% GDP) −0.027 −0.063*−0.027 (0.029) (0.038) (0.030) Public expenditure on culture (% GDP) 0.016 0.118*** 0.016 (0.029) (0.032) (0.028) Individual control variables Age −0.120*** −0.135*** −0.120*** (0.006) (0.008) (0.006) Squared age 0.212*** 0.237*** 0.212*** (0.004) (0.006) (0.004) Sex—Man −0.103*** −0.077*** −0.103*** (0.007) (0.009) (0.007) Political preferences—Left −0.147*** −0.145*** −0.148*** (0.008) (0.010) (0.008) Political preferences—Right 0.136*** 0.164*** 0.137*** (0.008) (0.011) (0.008) Household size—2 0.130*** 0.114*** 0.130*** (0.012) (0.016) (0.011) Household size—3 0.086*** 0.086*** 0.086*** (0.013) (0.017) (0.013) Household size—4 or more 0.124*** 0.125*** 0.124*** (0.013) (0.017) (0.013) Marital status—Living with partner −0.205*** −0.238*** −0.205*** (0.012) (0.016) (0.012) (Continues) (8) (9) (10) Marital status—Single −0.417*** −0.439*** −0.418*** (0.012) (0.016) (0.012) Marital status— Separated or divorced −0.653*** −0.653*** −0.653*** (0.014) (0.018) (0.014) Marital status—Widowed −0.493*** −0.491*** −0.493*** (0.015) (0.020) (0.015) Age education—16–19 years 0.203*** 0.240*** 0.204*** (0.010) (0.013) (0.010) Age education—More than 20 years 0.513*** 0.550*** 0.514*** (0.011) (0.015) (0.011) Age education—Still studying 0.034 0.024 0.035 (0.074) (0.107) (0.080) Occupation—Managers 0.196*** 0.157*** 0.196*** (0.015) (0.021) (0.015) Occupation—Other white collars −0.062*** −0.114*** −0.062*** (0.015) (0.021) (0.015) Occupation—Manual workers −0.236*** −0.290*** −0.236*** (0.014) (0.019) (0.013) Occupation—House person −0.251*** −0.307*** −0.251*** (0.017) (0.025) (0.017) Occupation— Unemployed −1.140*** −1.141*** −1.141*** (0.017) (0.024) (0.017) Occupation—Retired −0.266*** −0.318*** −0.267*** (0.015) (0.021) (0.015) Occupation—Student 0.429*** 0.478*** 0.428*** (0.075) (0.109) (0.080) Note: Standard errors in parentheses, observations nested at country–year level. Source: Own elaboration based on European Commission(2020), International Monetary Fund(2024), and World Bank(2024). *p < 0.1. **p < 0.05. ***p < 0.01. TABLE A3 | (Continued) 14676435, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/kykl.12466 by Readcube (Labtiva Inc.), Wiley Online Library on [23/05/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License