Spending on Education, Human Capital, and Economic Growth in Central America: A Panel Data Analysis with Driscoll-Kraay Standard Errors

Abstract

The main purpose of this study is to assess the effect of current public expenditure on education and human capital on economic growth in Central America between 1992 and 2021. In this context, data on education spending and human capital for Guatemala, Honduras, El Salvador, Nicaragua, and Costa Rica were analyzed using a panel data approach with Driscoll–Kraay standard errors. The time series were primarily obtained from the online databases of the World Bank, UNESCO, and national public sources. The results show a positive and significant effect of current education expenditure and human capital formation on the economic growth of Central America. This research provides empirical evidence on a topic that has been scarcely examined in the Central American regional context, and its findings constitute relevant input for scholars, practitioners, and policymakers.

1. Introduction

The historical trajectory of Guatemala, El Salvador, Honduras, Nicaragua, and Costa Rica is marked by fragmented political evolution and a dependent economic structure. During the Spanish colonial period, these territories belonged to the General Captaincy of Guatemala and were integrated into an extractive economic system based on the colonial labor system known as “Encomienda” and agricultural monocropping, which consolidated unequal social structures dominated by landowner elites (Bulmer-Thomas, 1987). After independence in 1821, they attempted a common political project through the Federal Republic of Central America (1823–1838), but institutional instability and conflicts between liberal and conservative factions led to its dissolution (Woodward, 1999). Each country then followed its own path under primary-export economies that were highly vulnerable to external demand and the influence of transnational corporations such as the United Fruit Company (Bulmer-Thomas, 2003; Colby, 2011). By the late twentieth century, the integrationist ideal re-emerged with the creation of SICA in 1991, although structural gaps in inequality and institutional weakness persisted (CEPAL, 2018). In 2004, the five Central American countries and the Dominican Republic signed a free trade agreement known as CAFTA-DR (Dunoff, 2010). The initiative generated measurable welfare gains and improvements in real income for several member countries, although with asymmetric effects across them. Overall, the agreement strengthened regional trade integration, but it did not translate uniformly into reduced inequality or sustained employment growth (Rojas Rodríguez & Matschke, 2023).

Central America shares a common historical trajectory and several structural constraints, including small economic size, high external dependence, and persistent inequality. However, the region is characterized by marked internal heterogeneity. Costa Rica stands out due to stronger state capacity, more consolidated institutions, and the development of a relatively broad welfare state, which has translated into superior social outcomes. In contrast, Guatemala, Honduras, El Salvador, and Nicaragua continue to exhibit weaker state structures, less consolidated democratic systems, and higher levels of social violence (Afşar & Sanchez, 2024; Díaz Arias, 2020; Izarra & Delgado, 2020). Productive integration within global value chains is likewise uneven: Costa Rica has specialized in higher-technology manufacturing and knowledge-intensive services, El Salvador in the apparel industry, and the remaining countries remain largely concentrated in primary activities and low-technology sectors (Delgado & Rogel, 2025). These disparities are reflected in competitiveness, science, and innovation performance, with Costa Rica leading the region and the other countries displaying low investment levels and still-fragile policy frameworks (Melara-Gálvez & Morales-Fernández, 2022; Viales-Hurtado et al., 2021).

In this context, the following question arises: To what extent does current spending on education and human capital development translate into productivity and economic growth in Central America? The relationship between spending on education, human capital, and Gross Domestic Product (GDP) growth in Central America is complex and multidimensional. Most empirical studies indicate that, while increased investment in education can positively influence economic growth, the effectiveness of such investments often depends on the allocation and quality of educational resources, the country’s context, and income level.

Human capital theories (Becker, 1964; Schultz, 1961) and endogenous growth theory (Lucas, 1988; Romer, 1990) provide a central analytical framework for explaining sustained economic growth, as they emphasize the role of internal factors in long-term expansion. From this perspective, Lucas (1988) argues that human capital functions as a core driver of continuous economic development. In this framework, knowledge not only accumulates progressively, but also reshapes productive structures, generating positive externalities that enhance productivity beyond the direct effect of initial investments.

According to Blankenau et al. (2007), public education expenditure is a key determinant of sustained economic growth because it promotes human capital accumulation. However, this relationship is neither linear nor direct. Spending increases on education may be offset by fiscal effects that partially neutralize its benefits, producing ambiguous net impacts on economic growth (Blankenau et al., 2007). In this same line of thought, human capital, particularly that derived from higher education, shows a positive and statistically significant correlation with GDP growth, especially in economies that allocate educational resources efficiently (Judson, 1998; Yu et al., 2014). Empirical evidence suggests that targeted investments in education, especially at the tertiary level, generate higher economic returns than indiscriminate education spending (Yu et al., 2014). Nonetheless, efficiency in the use of public education resources is crucial, since inadequate budget allocation can substantially diminish the positive effects of this investment on economic growth (Judson, 1998).

In Latin America, and particularly in the Central American region, empirical evidence on the relationship between investment in education, human capital, and economic growth is limited and shows heterogeneous results. Some studies have identified positive effects of public spending on education and health on economic growth in Latin America (Ramirez & Nazmi, 2003). However, more recent research indicates that this relationship is not universal. In the specific case of Honduras, no statistically significant association was found between education expenditure and economic growth (Villela & Paredes, 2022). Several authors also agree that the volume of investment is not the only factor that matters. Its effectiveness is equally important. In this regard, the quality of the education system is recognized as a key determinant of economic growth and of efforts to strengthen regional integration processes in Central America (Cáceres, 2021; Sorto-Bueso & Paredes Heller, 2023).

Although prior studies have explored the links between education spending, human capital, and economic growth in Central America (Loening et al., 2010; Sorto-Bueso & Paredes Heller, 2023; Villela & Paredes, 2022), significant gaps persist. Conceptually and empirically, it remains unclear how education expenditure translates into human capital accumulation and economic growth, especially when alternative spending measures and econometric methods capturing intra-country dynamics are applied. These limitations are amplified in developing economies with heterogeneous institutions, leaving regional evidence fragmented and methodologically diverse, which restricts a comprehensive understanding of the mechanisms through which education investments drive human capital and economic development.

Against this backdrop, structural asymmetries, recurrent episodes of political instability, and heterogeneous education reforms have generated divergent trajectories in educational investment and human capital accumulation across Central America. These conditions highlight the relevance of a comparative econometric approach capable of identifying robust empirical regularities that extend beyond country-specific institutional arrangements and historical dynamics.

This article examines the empirical relationship between current education expenditure, human capital, and economic growth in the Central American countries: Guatemala, El Salvador, Honduras, Nicaragua, and Costa Rica, which are currently part of the Dominican Republic–Central America–United States Free Trade Agreement (DR-CAFTA) (CEPAL, 2007).

2. Literature Review

Achieving sustainable economic growth is one of the most significant challenges for Central American countries, a region marked by deep inequalities in education, health, and technology (Ulku & Zaourak, 2021). In this context, human capital formation and public investment in education emerge as key factors for promoting regional development.

There is broad academic consensus that education spending and human capital are central drivers of economic growth and poverty reduction (Collin & Weil, 2020; Goczek et al., 2021; Li et al., 2024; Liao et al., 2019; Lutz, 2009; Maneejuk & Yamaka, 2021; Yu et al., 2014). The study by Phale et al. (2021), using multivariate panel data, showed that adjusted savings for education expenditure, tertiary enrollment, scientific and technical publications, and mobile phone subscriptions positively influence economic growth in the SADD region (16 African countries). Similarly, Fathy and Lenar (2018), applying panel data analysis for 15 MENA countries, demonstrated through cointegration a long-run relationship between human capital and GDP per capita. Variables such as tertiary education, labor participation, unemployment, and life expectancy are decisive. However, other studies show that the positive effect is not universal. More developed countries or regions benefit more from this effect than developing ones (Bayır & Zengin, 2024; Blankenau et al., 2007).

In contrast, Almutairi (2024) using an autoregressive distributed lag (ARDL) model for the period 1990–2018, found that both the gross tertiary enrollment ratio and scholarships, used as proxies for human capital, are negatively and significantly associated with economic growth in Saudi Arabia. Eggon et al. (2015), employing traditional cross-sectional techniques and dynamic panel models in a sample of 49 African countries between 1996 and 2010, found that public spending on education and health has a negative impact on economic growth, whereas human capital stock indicators show a slight positive effect.

Several studies have examined the link between education and economic growth in Latin America. Kiran (2014) analyzed 18 Latin American countries between 1970 and 2009 and found a cointegrating relationship between education expenditure and economic growth, except in Chile, Guyana, Jamaica, Nicaragua, Paraguay, Peru, and Uruguay. The study highlights the importance of education in driving GDP growth. Cerquera Losada et al. (2022) used a fixed-effects panel data model for eight South American countries between 2003 and 2018 and concluded that public spending on education maintains a positive and significant relationship with per capita income. Likewise, Osiobe (2020) applied panel models and Granger causality tests to 14 Latin American countries from 1970 to 2014—including Costa Rica, Honduras, El Salvador, and Nicaragua, and found that a 1% increase in human capital, measured in years of schooling, is associated with a 0.21% increase in real GDP per capita. However, the study did not find Granger causality from human capital to economic growth.

Villela and Paredes (2022) analyzed the relationship between public education spending, human capital, and economic growth in Honduras during 1990–2020 using instrumental variables with World Bank data. Their results show no significant correlation between education expenditure and economic growth, and they also find that human capital does not contribute meaningfully to growth in that period. In contrast, Zuniga Figueroa et al. (2018), applied a log-log elasticity model to six Central American countries (Guatemala, El Salvador, Honduras, Nicaragua, Costa Rica, and Panama) over 26 years and concluded that education has a positive effect on GDP per capita. Public spending on education and illiteracy rates were the variables with the strongest statistical influence.

In summary, the literature on the relationship between education expenditure, human capital, and economic growth can be broadly organized into three strands. A first group of studies reports positive and long-run effects, portraying education investment as a central driver of economic development, both at the global level (Fathy & Lenar, 2018; Phale et al., 2021) and within Latin America (Cerquera Losada et al., 2022; Kiran, 2014; Osiobe, 2020; Zuniga Figueroa et al., 2018). In contrast, a second strand records weak, non-significant, or even negative effects, often attributed to institutional inefficiencies, misallocation of resources, or limited productive absorption of human capital, as evidenced in diverse contexts such as Saudi Arabia (Almutairi, 2024), Africa (Eggon et al., 2015), and Honduras (Villela & Paredes, 2022). A third, more recent body of work advances conditional findings, arguing that the growth payoff of education depends on the level of development and, critically, on the transition from educational quantity to quality (Bayır & Zengin, 2024; Blankenau et al., 2007).

Despite the prevalence of studies reporting positive associations, empirical evidence remains mixed, particularly in developing economies. These divergent results indicate that the effectiveness of education spending depends on structural and institutional conditions and on the capacity to translate educational attainment into productivity gains. In Central America, marked by persistent structural constraints and institutional heterogeneity, the direction and magnitude of these effects remain an open empirical question, motivating the need for rigorous econometric analysis.

3. Methodology and Data

3.1. Research Question and Hypothesis

The methodological design focused on addressing the following question: How is the relationship between economic growth, education expenditure, and human capital in Central American countries during the period of 1992–2021? Based on the contribution of Mankiw et al. (1992) who extend the classical Solow model by incorporating human capital as an explicit factor of production, the research hypothesis was formulated as follows: There is a positive and significant effect between economic growth, current education expenditure, and human capital in Central American countries.

3.2. Approach and Model

This study followed a quantitative, non-experimental, retrospective longitudinal approach, using an econometric panel data model, as in the works of Phale et al. (2021) and Cerquera Losada et al. (2022). Panel data analysis allows for simultaneous consideration of cross-country differences and temporal changes, improving estimation accuracy and reducing biases associated with unobservable country-specific factors. The model is based on a long panel with five groups and thirty time units and takes the following form:

$$Y_{it}=\beta X_{it}+{\alpha }_i+{\epsilon }_{it}$$

(1)

In this model, Y is the dependent variable; X is the independent variable; i denotes the individual; t denotes time; β is the slope; α_i represents individual effects; and ${\epsilon }_{it}$ is the error term (Baltagi, 2021).

Based on this structure, the panel-data equation for this study was specified as follows:

$${Lgdp\_pc}_{it}={\beta }_0+{\beta }_1{Inflation}_{it}+{\beta }_2{Lsav\_edu}_{it}+{\beta }_3{Ter\_sch\_enr}_{it}+{\eta }_i+{\mu }_{it}$$

(2)

In this equation, ${Lgdp\_pc}_{it}$ is the natural logarithm of GDP per capita (current USD) and represents the region’s economic growth; ${Inflation}_{it}$ is inflation measured as the GDP deflator (annual %, used as a control variable); ${Lsav\_edu}_{it}$ is the natural logarithm of Adjusted Savings: Education Expenditure (current USD), used as a proxy for current education spending (Bingöl, 2020; Chlebisz & Mierzejewski, 2020; Elom et al., 2024a, 2024b; Günay, 2025; Phale et al., 2021; uz Zaman et al., 2021). This indicator captures the set of current operating expenses required for the functioning of the education system, including staff salaries, but excluding capital investments in infrastructure and equipment (World Bank Group, 2025). While this measure does not strictly correspond to budgeted educational spending, its use is conceptually consistent with endogenous growth models, in which investment in human capital is a fundamental determinant of sustained economic growth (Lucas, 1988; Romer, 1990). In this sense, the variable captures the public effort aimed at developing productive capacities, rather than the accounting level of educational spending. ${Ter\_sch\_enr}_{it}$ is the indicator for tertiary school enrollment (gross %), used as a proxy for human capital, also following Phale et al. (2021). It corresponds to the total number of students enrolled in higher education, regardless of age, expressed as a percentage of the population in the age group five years after completing secondary education (World Bank Group, 2025); β_k are the coefficients (unknown parameters of interest); ${\eta }_i+{\mu }_{it}$ are the model residuals (Sorto-Bueso & Paredes Heller, 2023). The empirical model adopts a semi-logarithmic specification. Accordingly, the coefficients associated with the logarithmic variables can be interpreted as elasticities, while the coefficients of the variables expressed in levels should be interpreted as semi-elasticities.

A potential limitation of this study lies in the measurement of human capital, which uses enrollment rates as a proxy due to data constraints. While educational expansion shows a positive impact on economic growth, it captures quantity rather than quality. As highlighted in the literature, cognitive skills and learning outcomes are crucial drivers of growth (Cáceres, 2021; Goczek et al., 2021; Hanushek & Woessmann, 2012; Jamel et al., 2020). In Central America, long-term quality indicators such as PISA scores, standardized assessments, or R&D investment are limited. Thus, future research should integrate quality-adjusted human capital measures to refine growth estimates.

3.3. Population and Sample

This study constructed a panel for five countries: El Salvador, Guatemala, Honduras, Nicaragua, and Costa Rica. The sample does not include Panama, as it is not part of the CAFTA-DR Central American region and due to its unique economic structure within the region. Panama’s economy is characterized by a growth model predominantly oriented toward logistics, financial services, and international trade, supported by strategic assets such as the Panama Canal. This productive specialization generates a growth dynamic that is less dependent on domestic human capital and public education spending, and more linked to quasi-exogenous rents and foreign direct investment (Hausmann et al., 2017; World Bank Group, 2025).The time series for each variable covers the period from 1992 to 2021, equivalent to 30 years. The variables representing GDP per capita (Lgdp_pc) and Adjusted Savings: Education Expenditure (Lsav_edu) are complete for all five countries. For the control variable inflation (Inflation), one observation was removed because it was considered an outlier. The variable Tertiary School Enrollment (Ter_sch_enr) contains 11 missing values due to limitations in the databases. Consequently, the panel was configured as unbalanced (Baltagi, 2021). Descriptive statistics of the panel are presented in

Table 1. Descriptive statistics for the panel data.

Variable		Mean	Std. Dev.	Min	Max	Observations
	overall	7.7595	0.7132	5.9773	9.4638	N = 150
Lgdp_pc	between		0.6135	7.1478	8.7165	n = 5
	within		0.4533	6.5889	8.5068	T = 30
	overall	6.8673	5.0081	−5.6101	23.7140	N = 149
Inflation	between		2.4230	3.3724	9.1254	n = 5
	within		4.5103	−4.1807	22.8026	T-bar = 29.8
	overall	20.1130	1.0062	17.6751	22.1806	N = 150
Lsav_edu	between		0.6705	19.1633	20.9915	n = 5
	within		0.8065	18.4216	21.3796	T = 30
	overall	22.2699	11.7063	7.7101	61.5639	N = 139
Ter_sch_enr	between		10.7929	14.1957	39.7510	n = 5
	within		6.2737	8.5121	44.0827	T-bar = 27.8

Note: The table summarizes the descriptive statistics for the model variables, which serve as proxies for economic growth, the control variable (inflation), current education expenditure, and human capital.

.

3.4. Data Collection

Data collection relied on public sources (Banco Mundial (https://datos.bancomundial.org/indicador/SE.TER.ENRR, accessed on 22 December 2024), UNESCO (https://databrowser.uis.unesco.org/browser/EDUCATION/UIS-SDG4Monitoring/t4.3, accessed on 22 December 2024)) and national repositories. The variable Ter_sch_enr initially contained 56 gaps. A rigorous imputation process was implemented, completing 45 missing values through interpolation or linear regression, using an auxiliary national enrollment variable with statistically significant association. This procedure is supported by the literature (Moritz & Bartz-Beielstein, 2017; Mostafa, 2019) and reduced missing data to 7% in the critical variable and 2% in the full panel. A robustness check was conducted to evaluate the effects of imputing the variable Ter_sch_enr by comparing the imputed panel with the original dataset. The coefficient for tertiary enrollment remained stable, decreasing slightly from 0.0121 to 0.0104, while retaining statistical significance at the 10% level despite fewer degrees of freedom. Simultaneously, the coefficient for Lsav_edu increased marginally (0.4506 to 0.4565) and remained significant below the 1% level. The stability in magnitude and sign across specifications indicates that the imputation process does not introduce systematic bias, supporting the methodological validity of the imputed model. This effort demonstrates strong commitment to the integrity of the time series, which is essential for the validity of any longitudinal analysis.

The national repositories or secondary sources that were used are: the National Council of Rectors CONARE (https://repositorio.conare.ac.cr/home, accessed on 15 January 2025) in Costa Rica; the National Institute of Development Information INIDE (https://www.inide.gob.ni/Home/Anuarios, accessed on 15 January 2025) of Nicaragua; the report Behavior of the gender variable in higher education in Nicaragua in UNESDOC (https://unesdoc.unesco.org/ark:/48223/pf0000139978?posInSet=1&queryId=87872143-3b37-4eac-b3e4-b08a6b1897d6, accessed on 16 January 2025); the National Institute of Statistics INE (https://www.ine.gob.gt/educacion/, accessed on 16 January 2025) of Guatemala; and the article by Tobar Piril (2011). The imputation process reduced missing data to 7% for Ter_sch_enr and 2% for the entire panel (Moritz & Bartz-Beielstein, 2017).

The databases were organized by indicator and country, and the tables were downloaded in Excel format. The data was then structured for import and processing in STATA 14. Although the time series were not fully complete for all countries, this did not constrain the panel configuration because the proportion of missing data remained below 5% (Baltagi, 2021).

It is important to acknowledge that the relationship between economic growth, education expenditure, and human capital is potentially inverse or bidirectional (Osiobe, 2020; Yu et al., 2014). This feature introduces an inherent risk of endogeneity associated with simultaneity and reverse causality. Although there are econometric techniques specifically designed to address this issue, such as instrumental variable estimation or the Generalized Method of Moments (GMM) in dynamic panels (Blundell & Bond, 1998), their application requires the availability of valid instruments. Consequently, the results should be interpreted as robust economic associations rather than as strict estimates of causal effects.

3.5. Analytical Approach

Given the magnitude of the indicators GDP per capita and Adjusted Savings: Education Expenditure, expressed in billions of dollars, a logarithmic transformation was applied (Phale et al., 2021; Sorto-Bueso & Paredes Heller, 2023). This statistical procedure serves a dual purpose. It standardizes the scale of the measurements by removing dependence on the original units. It also mitigates the effect of large year-to-year fluctuations, resulting in a more stable and manageable time series for analysis (Box & Cox, 1964). The preliminary statistical analysis included an evaluation of correlation and collinearity among the selected variables that would form the econometric model (Gujarati & Porter, 2010). Several diagnostic tests were then conducted to validate the model specifications. The Hausman and Mundlak tests were applied to justify the use of a fixed-effects methodology over random effects. The fixed-effects model, by incorporating country-specific intercepts, removes potential bias arising from the correlation between unobserved effects and the regressors. The Mundlak test was particularly relevant. According to Clark and Linzer (2015), the power of the Hausman test may be weakened in settings with small N and non-spherical errors (heteroskedasticity and autocorrelation). Therefore, the choice between fixed-effects (FE) and random-effects (RE) should also rely on substantial criteria and complementary diagnostics. Even if Hausman does not reject the consistency of RE, in macroeconomic panels with geographically close countries it is highly plausible that unobserved country-level factors, such as institutions, government quality, social norms, and administrative capacity, are correlated with the regressors of interest (e.g., education spending and enrollment rates). Under such correlation, RE becomes inconsistent, while fixed-effects estimates remain valid (Mundlak, 1978). The modified Wald test was then applied to verify the hypothesis of homoskedasticity (Baum & College, 2001), and the Wooldridge test was used to confirm the absence of autocorrelation in the model errors (Gujarati & Porter, 2010). Finally, the Breusch-Pagan Lagrange Multiplier test was performed to assess cross-sectional dependence (Breusch & Pagan, 1980).

The characteristics of the panel and the test results enabled the identification of the appropriate panel model for estimating the coefficients of Equation (2). According to Tugcu (2018) and Hoechle (2007), the Driscoll-Kraay model is the most suitable for panels with few units (N) and long time series (T). This estimator is robust to heteroskedasticity, autocorrelation, and general cross-sectional dependence, as also applied by Paul et al. (2021) and Ewane (2023).

3.6. Ethical Considerations

The data used in this study comes from public and open-access sources, such as the World Bank, UNESCO, and national repositories in Central America. The imputation of missing values in the time series for the indicator Tertiary School Enrollment (% gross) was carried out following scientific integrity and methodological consistency criteria (Martínez-Plumed et al., 2019). Because some countries showed information gaps in the World Bank records, official national sources on total higher-education enrollment were consulted, and a statistically significant relationship between the two series was established. Based on this relationship, linear estimation methods were applied to complete the missing values, ensuring coherence and continuity in the time series. From an ethical standpoint, this procedure is grounded in transparency and data traceability. The imputations were clearly documented, differentiated from observed values, and used solely for analytical purposes, avoiding any manipulation that could distort the results or compromise the validity of the conclusions.

4. Results and Discussion

4.1. Results

This section presents the findings of the panel data analysis for Central America. Figure 1 displays the time series of the model’s variables: the natural logarithm of GDP per capita (current USD), the natural logarithm of Adjusted Savings: Education Expenditure (current USD), tertiary school enrollment rate, and inflation, which is included in the model as a control variable.

Table 1 presents the descriptive statistics of the dataset. The panel includes five countries and 150 temporal observations for the variables Lgdp_pc and Lsav_edu, and 149 observations for inflation due to the exclusion of one outlier. The variable Ter_sch_enr contains 139 observations, most of them missing in Nicaragua and Honduras, resulting in an unbalanced panel (Baltagi, 2021). It is noteworthy that this variable exhibits a relatively high standard deviation. This is explained by Costa Rica’s considerably higher tertiary enrollment rates compared with its neighbors. For example, in 2021, slightly more than 63% of Costa Rican youth of tertiary-school age were enrolled, while only 30% were enrolled in El Salvador and Guatemala. In Honduras and Nicaragua, the rate averaged 23%. It is also relevant to highlight that Costa Rica’s GDP per capita is nearly three times that of El Salvador and Guatemala, five times that of Honduras, and six times that of Nicaragua. A similar pattern emerges for Adjusted Savings: Education Expenditure. However, for these two indicators, the logarithmic transformation proved suitable, as their standard deviations, between, within, and overall, are small relative to their means (Phale et al., 2021).

Table 2. Variance Inflation Factor (VIF).

Variable	VIF	1/VIF
Inflation	1.40	0.7135
Lsav_edu	2.75	0.3633
Ter_sch_enr	2.18	0.4596
Mean VIF	2.11

Note: The table reports the Variance Inflation Factor (VIF) and the tolerance values (1/VIF) for the model’s independent variables, confirming a low risk of multicollinearity.

presents the Variance Inflation Factor (VIF), which assesses how rapidly variances and covariances increase while isolating the combined effect of the predictors on each variable (Gujarati & Porter, 2010). The results show VIF values below 5 and tolerance levels (1/VIF) above 0.2, indicating a low risk of multicollinearity in the model (Miles, 2014).

Once the suitability of the variables for panel-data analysis was confirmed, the Hausman and Mundlak tests were applied to determine whether a FE or RE specification was appropriate. The Hausman test was not significant, indicating that the RE model was more efficient (Gujarati & Porter, 2010). However, the Mundlak test reported a statistic of 68.73, significant at the 0.1% level, which invalidates the use of the RE model. In practice, the Mundlak and Correlated Random Effects (CRE) approach, i.e., including the country-level means of the regressors in an RE-type specification, allows researchers to “audit” the key assumption underlying RE. The joint significance of these terms provides evidence of correlation between x_it–α_i, thereby supporting the use of FE (or CRE) even when the Hausman test is not significant (Mundlak, 1978).

Table 3. Hausman and Mundlak Tests.

Test	Chi2	p-Value	Condition
Hausman test	2.82	0.4206	RE better than FE
Mundlak test (robust)	68.73	0.0000	Use FE/CRE instead RE

Note: The table presents the results of the Hausman and Mundlak tests. Although the Hausman test is not significant and suggests a preference for a random-effects model, the Mundlak test confirms the appropriateness of using a fixed-effects model or a correlated random-effects specification rather than a standard random-effects model.

reports the results of both tests.

Table 4. Summary of Diagnostic Tests.

Test	Chi2	F(1,4)	p-Value	Condition
Modified Wald test	73.39	---	0.0000	heteroskedastic
Wooldridge test	----	65.14	0.0013	first-order autocorrelation
Breusch-Pagan LM test	23.78	----	0.0082	cross sectional dependence

Note: The table summarizes the results of the assumption tests for the model variables, including homoscedasticity, autocorrelation, and cross-sectional dependence.

reports the results of the modified Wald test strongly rejects the null hypothesis of homoscedasticity (χ² = 73.39, p < 0.001), indicating the presence of groupwise heteroskedasticity across countries (Baum & College, 2001). The Wooldridge test confirms first-order autocorrelation, with an F-statistic of 65.14 significant at the 1% level (Gujarati & Porter, 2010). In addition, the Breusch–Pagan Lagrange Multiplier test reveals significant cross-sectional dependence (χ² = 23.78, p < 0.01), suggesting contemporaneous correlation of error terms among the five Central American economies (Breusch & Pagan, 1980). A summary of these results is provided in Table 4.

These findings indicate that conventional fixed-effects or random-effects estimators would yield inconsistent standard errors and unreliable inference (Beck & Katz, 1995). Groupwise heteroskedasticity and serial correlation reflect structural cross-country differences and the temporal persistence typical of macroeconomic data, while cross-sectional dependence highlights the relevance of common regional shocks. Consequently, the Driscoll–Kraay estimator is employed, as it provides robust inference in the presence of heteroskedasticity, autocorrelation, and general forms of cross-sectional dependence (Hoechle, 2007). The model results are summarized in

Table 5. Driscoll-Kraay Model results.

Variable	Coef.	Std. Err.	T	P > z	[95% Conf. Interval]
Lsav_edu	0.45062	0.04532	9.94	0.000	[0.3579, 0.5433]
Ter_sch_enr	0.01217	0.00525	2.32	0.028	[0.0014, 0.0229]
Inflation	−0.00415	0.00564	−0.74	0.468	[−0.0157, 0.0074]
constant	−1.5254	0.8185	−1.86	0.073	[−3.1993, 0.1486]
Prob > F	0.000
R2	0.956

Note: The table reports the results of the Driscoll–Kraay model. The coefficients for the proxies representing current expenditure on education (Lsav_edu) and human capital (Ter_sch_enr) are positive and statistically significant at levels below 5%. The inflation variable (included as a control) displays a negative coefficient, although it is not statistically significant.

.

The high R² value should be interpreted with caution. In macroeconomic panel models with fixed effects and highly persistent variables, elevated R² values are common. They mainly reflect long-run within-country variation rather than the predictive power of the model (Baltagi, 2021; Wooldridge, 2010). In addition, the presence of common trends and shared shocks may increase goodness of fit without implying overfitting (Gujarati & Porter, 2010). Therefore, R² is not the primary criterion for model validation in this context. Robust inference constitutes the central objective of the adopted empirical approach. The F-statistic is highly significant (p < 0.001), confirming that the set of explanatory variables has a statistically significant joint effect on the dependent variable Lgdp_pc, even under the robust Driscoll–Kraay correction, which accounts for heteroskedasticity, autocorrelation, and cross-sectional dependence in the panel. The Driscoll-Kraay estimator implemented through the xtscc command in STATA selected a maximum lag of two periods (“maximum lag = 2”) for the correction of serial autocorrelation. This means that the variance–covariance matrix was adjusted considering temporal dependencies of up to two periods, thereby strengthening the robustness of the standard errors reported (Hoechle, 2007).

The coefficients for Lgdp_edu and Ter_sch_enr are positive and statistically significant at the 0.1% and 5% levels, respectively. The coefficient of the inflation variable, although negative, is not statistically significant. Based on these estimates, the coefficients can be expressed as follows: β₀ = −1.53 (p < 0.1), β₁ = 0.45 (p < 0.01), β₂ = 0.0122 (p < 0.05) and β₃ = −0.0042 (not significant). Accordingly, the econometric model is expressed as in Equation (3).

Lgdp_pc_it = −1.53 + 0.45Lsav_edu_it + 0.0122Ter_sch_enr_it − 0.0042Inflation_it + η_i + μ_it

(3)

4.2. Discussion

The coefficient associated with Adjusted Savings: Education Expenditure, used as a proxy for current operational spending in this sector, was 0.45062 with a p-value below 0.01, indicating statistical significance. Under the logarithmic transformation, a one-point increase in this indicator is linked to an approximate 0.45% rise in GDP per capita, holding all other variables constant. This finding aligns with the evidence reported by Çetin (2023) for the United States and Phale et al. (2021) for Southern Africa, confirming that the relationship between educational spending and economic growth is generally positive and statistically significant. Although the results confirm the positive link, they suggest that the marginal impact of education expenditure in Central America is very weak. This outcome is consistent with regional studies for South America (Cerquera Losada et al., 2022) and the Northern Triangle (Sorto-Bueso & Paredes Heller, 2023), both of which describe a limited effectiveness of educational investment in these economies. This result should be interpreted considering that education spending reflects a measure of investment in human capital from a fiscal and economic sustainability perspective, and not a strict measurement of educational expenditure. Therefore, the estimated coefficient represents the impact of public education efforts in terms of human capital accumulation, rather than a direct budgetary effect. Such low efficiency reflects the fact that in countries like Guatemala, El Salvador, and Honduras, a large share of spending goes to wages and inefficient operational outlays, reducing the productive contribution of education to growth, as suggested by Urhie (2014) and Villela and Paredes (2022). These findings raise fundamental questions about the effectiveness of current public policies and underscore the need to examine whether educational resources are being managed in ways that maximize their contribution to growth.

The coefficient for tertiary-level school enrollment, used as a proxy for human capital, was positive (0.01217) with a p-value below 0.05, confirming statistical significance. Since this coefficient corresponds to a variable measured in levels within a semi-logarithmic specification, it should be interpreted as a semi-elasticity. A one–percentage-point increase in tertiary enrollment is associated with an approximate 1.22% increase in GDP per capita, for marginal changes. This finding is consistent with human capital theory, which argues that broader access to higher education strengthens labor productivity and thus fosters economic growth (Essardi & Razzouk, 2017). Although this study focuses on the upper levels of the education system, the literature suggests that early childhood development, particularly preschool education, may be a critical component for enhancing human-capital returns in later stages (Heckman & Masterov, 2007). Within this framework, the positive but marginal coefficient of tertiary enrollment rate indicates that expanding access to higher education, by itself, does not generate a strong impact on per-capita GDP growth. This pattern is consistent with studies reporting reduced effects in small and open economies (Martinez, 2019). Thus, the evidence suggests that the effectiveness of education spending depends not only on coverage but also on the quality and relevance of training. As emphasized by Krueger and Lindahl (2001), countries where education strongly drives growth tend to have highly consolidated systems.

Finally, the coefficient associated with the control variable Inflation was negative, with a p-value indicating no statistically significant effect on GDP per capita in this model. Nonetheless, the negative sign is consistent with the findings of Moreno-Brid et al. (2014) and Rivas and Badebuena (2008) confirming the internal coherence of the model.

Regarding the research gap, this study provides empirical evidence on the relationship between current educational expenditure, human capital development, and economic growth in Central America. The analysis addresses the scarcity of regional studies in the academic literature, and its findings are essential for understanding the joint impact of these variables on the region’s economy. Ultimately, it reinforces the notion that investment in education is a key driver of economic development. This study employed a panel data analysis using a Driscoll–Kraay model, correcting for autocorrelation, heteroskedasticity, and cross-sectional dependence. Using time series data from 1992 to 2021, this study identified a positive and statistically significant, though very weak, relationship among the variables. This methodological approach enabled the research question to be answered and this study’s main objective to be achieved.

With respect to the research hypothesis, the econometric results confirm it. The model reports a coefficient of +0.451 for the variable associated with current educational expenditure and +0.012 for the variable associated with human capital, both statistically significant (p < 0.05). These results indicate a positive and meaningful effect of educational spending and human capital on economic growth in Central America.

Altogether, the results indicate that education plays a necessary yet structurally constrained role in Central America’s growth process. Public education spending contributes positively to output; however, its limited impact reflects an expenditure structure dominated by recurrent spending rather than productivity-enhancing investment. At the same time, the modest effect of tertiary enrollment suggests that expanding educational coverage alone is insufficient to generate substantial growth effects in the absence of complementary conditions, including productive absorption capacity, technological upgrading, and institutional quality. Overall, the findings are consistent with a growth regime where human capital accumulation is only weakly linked to productivity gains.

Several limitations must be noted. The absence of complete time series for current educational expenditure across most Central American countries required the use of Adjusted Savings: Education Expenditure as a proxy variable. Although, according to the World Bank, this indicator is derived from the former, the coefficients obtained could vary depending on which measure is used. Likewise, this study focused exclusively on tertiary enrollment due to the lack of consistent information on university graduates, patent counts, or spending on research and development (R&D) in the region. This data scarcity limits the incorporation of other relevant components of human capital and innovation that would have contributed to a broader and more robust analysis. Another point worth highlighting is that the estimated coefficients reflect statistically significant relationships between the educational variables and economic growth. However, given the potential presence of endogeneity, these results should not be interpreted as direct causal evidence. In line with the empirical literature on economic growth based on macroeconomic panel data, the analysis is aimed at identifying patterns of association consistent with human capital theory, rather than establishing definitive causal mechanisms (Wooldridge, 2010). Although the model provides meaningful insights, it abstracts from time fixed effects and country-specific trends. This condition implies that certain global shocks or common trends may not be fully captured. Future studies could extend this work by incorporating more granular fixed-effects structures to better isolate the underlying causal mechanisms. Despite these constraints, the findings suggest that strengthening educational investment and expanding access to higher education are strategic pillars for promoting economic growth in Central America. However, fully leveraging human capital requires going beyond current expenditures and incorporating measures aimed at improving educational infrastructure and technologies, raising quality standards, and promoting public policies that encourage innovation and scientific-technological development.

Beyond statistical significance, the findings indicate that the effect of current education spending on economic growth in Central America may operate primarily through long-term, indirect channels, conditioned by structural constraints. In economies characterized by high labor informality, limited absorption of human capital in the formal sector, and low technological intensity, the macroeconomic return on human capital depends on the capacity of the institutional framework and the productive structure to translate educational attainment into labor efficiency. The estimated effect aligns with a context in which education is a necessary, but not sufficient, condition for growth, helping to explain both the heterogeneity observed across Central American countries and the mixed empirical results documented in the literature.

Future research should deepen the analysis in three key areas. First, potential endogeneity in educational spending should be examined using Instrumental Variables or the Generalized Method of Moments (GMM) to establish a more robust causal relationship between education and economic growth, even disaggregated by country. Second, the efficiency of educational spending should be assessed through frontier-based approaches such as Data Envelopment Analysis (DEA) or Stochastic Frontier Analysis (SFA), which can identify regional disparities and support public policy decisions. Finally, human capital should be disaggregated using indicators of educational quality, such as graduation rates or labor-market insertion, to more accurately capture its true impact on innovation and sustainable economic development.

5. Conclusions

The findings of this study confirm a positive and statistically significant effect of current educational expenditure and human capital formation on economic growth in Central American countries.

Regarding current spending on education, the results show that a one-point increase in the adjusted-savings indicator allocated to this sector is associated with an increase of approximately 0.45% in GDP per capita. Although the impact is moderate, the evidence supports the hypothesis that investment in education contributes to economic development, particularly when understood as a long-term mechanism that promotes capacity accumulation and institutional strengthening. One limitation of this study lies in the use of the variable Adjusted Savings: Education expenditure, which, although widely used in the literature on sustainability and economic growth, is not fully comparable with conventional measurements of education expenditure, which were scarce or incomplete at the time this research was conducted. Nevertheless, its selection is justified by its ability to reflect public investment in human capital from a long-term perspective. Future research could complement this approach by incorporating traditional measures of education expenditure to assess the robustness of the results.

The human capital variable, proxied by tertiary enrollment, also exhibits a positive and statistically significant effect. In this case, a one-point increase in university enrollment rate translates into an approximate 1.22% rise in GDP per capita. This result underscores the importance of expanding access to higher education as a means of boosting productivity and sustaining long-term growth in the region, as well as guiding public policies focused on university quality and investment in R&D. From a methodological perspective, estimation through the Driscoll–Kraay panel model allowed this study to correct for autocorrelation, heteroskedasticity, and cross-sectional dependence, thereby ensuring the robustness of the results. This approach constitutes a valuable tool not only for the analysis of Central America but also for other regions facing similar structural challenges.

A relevant limitation of this study is the potential risk of endogeneity arising from the bidirectional or reverse relationship between economic growth and educational variables. Although this study does not implement dynamic estimation techniques such as the Generalized Method of Moments (GMM) or instrumental variables, it is acknowledged that these methodologies could strengthen causal identification in future research. In this regard, the results should be interpreted as empirical evidence of economic associations consistent with theory, rather than as strict causal estimates.

Overall, the results reinforce the relevance of education as a driver of economic growth. The empirical evidence presented offers essential input for designing public policies aimed at improving the efficiency of educational spending and strengthening access to higher education, with the goal of fostering sustainable economic growth in Central America.

This study stands out for its focus on Central America, assessing how current educational expenditure and human capital, measured through higher-education enrollment, affect economic growth. Its originality lies in addressing a region with specific structural challenges and in providing empirical evidence on emerging economies that remain understudied in the literature. The results offer meaningful guidance for scholars, practitioners, and policymakers by indicating that investment in higher education is a relevant channel for promoting sustained growth, enhancing competitiveness, and raising productivity. Likewise, the expansion of university access appears as a potential mechanism for reducing inequality and improving social well-being.

It is recommended that the methodological framework employed here be replicated in other contexts to identify regional patterns and heterogeneities. Moreover, further studies are needed to design and implement concrete policies that increase the efficiency of educational spending and strengthen its capacity to drive development. Finally, this research provides a solid foundation for guiding public interventions that position education as a central engine of regional development.